GitHub Copilot Fusion GA: Routing, Billing and Admin Controls

GitHub made Copilot Fusion generally available on October 28, 2026. Instead of selecting one model, users choose Standard, High or Max effort and let GitHub route or orchestrate eligible models. The system may use one model, cascade from an efficient model to a stronger one, or ask one model to critique another.

The model picker now selects a routing system

Fusion combines ideas from Project HydraFusion and Copilot automatic model selection. GitHub says the router examines each prompt and can choose single, cascade or critique execution. That may reduce manual switching, but it also makes the exact model path variable across tasks.

Teams evaluating output should record the effort level, client, policy state and available model pool. A later run may take a different route as policies or the service change. Reproducibility requires more context than a single model name.

GitHub’s benchmark comparison is not a universal cost result

GitHub reports Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of a named competitor, at roughly half the average API cost, half the tokens and 40% less time per task. Those are vendor-reported results on a particular index and configuration.

A Canadian development team should replay its own tasks and measure accepted changes, review time, regressions and total billed model use. A router can be efficient on average while escalating a difficult task to several charged models.

Each routed model can contribute to the bill

GitHub says customers are billed for each model Fusion uses at that model’s standard rate. Paid Copilot users receive a temporary 30% discount on model costs through Fusion, but the announcement does not specify a Canadian-dollar end date or guarantee a fixed cost per task.

Set budget alerts, compare effort levels and review model-attributed usage. Standard may be sufficient for routine work; Max should be justified by measured quality rather than left as an unquestioned default.

Administrators define the model pool

Fusion only uses models enabled by enterprise and organization policies. Administrators can separately control which approved models Fusion may select and can set an effort level as the managed default. That keeps central policy relevant even when the user does not select a provider directly.

Fusion is available on all Copilot plans in VS Code, the Copilot app, CLI and JetBrains, with other clients planned. Teams should update supported clients, verify policy inheritance and preserve human review for generated changes. Routing can coordinate models; it cannot assign accountability for code that ships.

Sources and editorial method

We use the linked original statements and reports to distinguish documented facts from our analysis. Company announcements describe their authors’ plans and claims; they are not independent verification of future outcomes. This article is dated and will be revised if material evidence changes. Read our editorial standards and corrections process.

GitHub: Route and orchestrate models with Copilot Fusion ↗ — Official October 28 GA announcement; routing, availability, billing and benchmark claims.